A GENETIC-BASED FUZZY LOGIC POWER SYSTEM STABILIZER FOR
MULTIMACHINE POWER SYSTEMS
M. A. Abido
Electrical Engineering Department
King Fahd University of Petroleum and Minerals
Dhahran 31261, Saudi Arabia
Abstract - This paper presents a novel approach to
combine genetic algorithms (GA) with fuzzy logic
systems to design a genetic-based fuzzy logic power
system stabilizer (GFLPSS) for multimachine power
systems. Incorporation of G A in fuzzy logic power
system stabilizers (FLPSSs) design will significantly
reduce the time consumed in the design process of
FLPSSs.It is shown in this paper that the performance
of FLPSS can be improved significantly by
incorporating a genetic-based learning mechanism. The
performance of the proposed GFLPSS under different
disturbances is investigated. The results show the
superiority of the proposed GFLPSS as compared to the
classical PSS and its capability to enhance system
damping to local as well as interarea modes of
oscillations. The capability of the proposed GFLPSS to
work cooperatively with the existing classical PSSs is
also demonstrated.
1. INTRODUCTION
Due to increasing complexity of electrical power
systems, there has been increasing interest in the
stabilization of such systems. In the past two decades,
the utilization of supplementary excitation control
signals for improving the dynamic stability of power
systems and damping out the low frequency oscillations
due to disturbances has received much attention [l-91.
Nowadays, the conventional power system stabilizer
(CPSS) - a fixed parameters lead-lag compensator is
widely used by power system utilities [3]. The gain
settings of these stabilizers are determined based on the
linearized model of the power system around a nominal
operating point to provide optimal performance at this
point. Generally, the power systems are highly nonlinear
and the operating conditions can vary over a wide range.
Therefore, CPSS performance is degraded whenever the
operating point changes from one to another because of
fixed parameters of the stabilizer.
Alternative controllers using adaptive control
algorithms have been proposed to overcome such
problems [5-61. However, most adaptive controllers are
designed on the basis of parameter identification of the
system model in real-time which results in time
consuming and computational burden.
0-7803-4053-1/97/$10.00 1997 IEEE
@
Authorized licensed use limited to: Sona College of Technology. Downloaded on August 02,2010 at 05:34:49 UTC from IEEE Xplore. Restrictions apply.
Y. L. Abdel-Magid
Recently, fuzzy logic power system stabilizers
(FLPSSs) have been proposed 17-91. FLPSSs appear to
be the most suitable stabilizers due to their lower
computation burden and robustness. Unlike the most
classical methods, an explicit mathematical model of the
system is not required to design a good FLPSS which
makes it more suitable for on-line computer control. In
addition, FLPSS can be easily set up using
microcomputer with A/D and D/Aconverters [101.
Although fuzzy logrc controllers showed promising
results, they are subjective and somewhat heuristic. In
addition, generation of membership functions, and the
choice of scaling factors are done either iteratively, by
trial-and-error, or by human experts. There is to-date no
generalized method for the formulation of fuzzy control
strategies, and design remains an ad hoc trial and error
exercise. That makes the design of fuzzy logic controller
a laborious and time-consuming task.
On the other hand, genetic algorithms (GA)are:
search algorithms based on the mechanics of naturd
selection and survival-of-the-fittest notion. The
principles of GA were first introduced by Holland in hs
pioneering work in the theoretical development and
adaptation in natural and amficial systems [ll].
Recently, GA have been applied to various power system
problems with promising results [12-141.
The recent approach is to integrate the use of GA
and fuzzy logic systems in order to combine their
different strengths [15-16]. A genetic-based fuzzy logic
- power system stabilizer for a single machine infinite bus
system has been proposed with promising results [171.
In this paper, we extend this approach to a
multimachine power system where the problem becomes
more realistic and complex. The results show that the
performance of FLPSSs can be significantly improved
by incorporating a genetic-based learning mechanism.
2. GENETIC ALGORITHMS
GA are exploratory search and optimization
procedures that were devised on the principles of natural
evolution and population genetics. Unlike other
optimization techniques, GA work with a population of
individuals represented by bit strings and m o w the
population with random search and competition. The
329
advantages of GA over other traditional optimization
techniques can be summarized as follows:
0 GA work on a coding of the parameters to be Step 5: Calculate R,(k) and eI(k)using
optimized, rather than the parameters themselves.
0 GA search the problem space using a population of
trials representing possible solutions to the problem,
not a single point, i.e. GA have implicit parallelism.
This property ensures GA to be less susceptible to
getting trapped on local minima.
0 GA use a performance index assessment to guide
the search in the problem space. Step 6: Compute the values of membership functions
0 GA use probabilistic rules to make decisions. N,{ej) and Pg{0j) defined as [7]
Typically, the GA starts with little or no knowledge
of the correct solution dependmg entirely on responses
from interacting environment and their evolution
operators to arrive at optimal or near optimal solutions.
In general, GA include operations such as reproduction,
crossover, and mutation. Reproduction is a process in
which a new generation of population is formed by
selecting the fittest individuals in the current population.
Crossover is the most dominant operator in GA. It is
responsible for producing new offsprings by selecting '0.0 V x I a
two strings and exchanging portions of their structures. 2[----12
x - a v x € 1a,b] (9)
The new offsprings may replace the weaker individuals @ ( x ; a , b , c )= c - a
in the population. Mutation is a local operator whch is 1 - 2[---]*
x - c v x E 3 b,c[
applied with a very low probability. Its function is to c - a
alter the value of a random position in a string. 1.0 v x 2 c

3. FUZZY LOGIC CONTROL SCHEME [17] @(x;c-b , -b~ 1 2 , ~VX

) Ic
Y(x;b,c)= (10)
l-@(~;c,c+b/2,c+b) VX >C
The stabilizing signal u is added to the excitation
loop as shown in Fig. 1. At time t, u(t) is given by

u(t) = U(k) ; kT,< t < (k+l)Ts (1)

The value of U(k) is determined at each sampling time

based on fuzzy logic through the following steps:
Step 1: Consider the jth machine where the stabilizer
will be installed.
Step 2: The speed deviation of thejth machine, do.@),
is measured at every sampling time, and the
acceleration of the machine, AJ{k),is calculated
bY
It is worth mentioning that these continuous
nonlinear membership functions are more
suitable for power system stability problem [7].
Step 7; Determine the value of the gain function G,(k)
defined as [XI

Step 8: Compute the stabilizing signal v ( k ) using

Step 3: Compute the scaled acceleration,A,fk), using

where Umarjis the maximum value of the

where Faj is the scaling factor for the jth
stabilizing signal at the jth machine.
machine acceleration.
Step 9: Consider another machine where the stabilizer
Step 4: The generator condition is given by the point
should be installed and go back to step 2.
G(k) where
Step 10: Increase k by 1 and return to step 1.

330

Authorized licensed use limited to: Sona College of Technology. Downloaded on August 02,2010 at 05:34:49 UTC from IEEE Xplore. Restrictions apply.
 The main tuning parameters of an FLPSS are Bv, size, maximum number of generations, and crossover
Faj , and Dj. For the optimal settings of these and mutation probabilities are chosen to be 30, 50, 0.75,
parameters, a quadratic performance index J is and 0.005 respectively. Fig. 3 shows the convergence
considered: rate of the performance index J.
N L
J = E[kT,Aoi(k)l2 initialization
i=l k=l

where N is the number of machines and L is the total

number of data points.
In the above index, the speed deviation of the ith
machme Awi(k) is weighted by the respective time kT,. time domain simulation of the system
and performance index evaluation
The index J is selected because it reflects small settling
time, small steady state error, and small overshoots. The +
tuning parameters are adjusted so as to minimize the GA operators
index J.

V I /-
External

Vr b

DIAMICROCOMFUTER AID A@
Fig. 2: Flow chart of the proposed GFLPSS design
Fig. 1: Study system configuration.

4. THE PROPOSED APPROACH

The major problem in the design of conventional

i 1.6E-2 4
FLPSSs is that the stabilizers are designed one at a time
and the parameters of each stabilizer are optimized
iteratively [8]. This process is time consuming and the
interaction between stabilizers is not taken into account.
This results in degradation of the stabilizer performance. 1.4E-2 , I I , I '
In the proposed approach, all stabilizers are designed 0.W 10.M 20.M 30.00 40.00 50.0
together and all parameters are optimized Generations
simultaneouslyusing GA as an optimization process. Fig. 3: Variation of the performance index J.
Applying the GA to the problem of PSS design
involves performing the following two steps. 5. SIMULATION RESULTS
1. The performance index value must be calculated for
each of the strings in the current population. To do In this study, the 10-machine 39-bus New England
this, the tuning parameters must be decoded from power system shown in Fig. 4 was considered [8]. Each
each string in the population and the system is machine has been represented by a fourth order two-axis
simulated to obtain the performance index value. nonlinear model. Generator G1 is an equivalent power
2. GA operations are applied to produce the next source representing parts of the U.S.-Canadian
generation of the strings. interconnection system. Details of the system data are
These two steps are repeated from generation to given in [8]. In this study, the following disturbances are
generation until the population converges producing an considered for the simulations:
optimal or near optimal parameter set. The
computational flow chart of the design process of the (a) Three phase fault for 0.10s at bus 29 at the end
proposed GFLPSS is shown in Fig. 2. of line 26-29.
The tuning parameters are coded in a binaq string. (b) Three phase fault for 0.15s at bus 15 at the end
The initial population is generated randomly. Population of line 14-15.

331

Authorized licensed use limited to: Sona College of Technology. Downloaded on August 02,2010 at 05:34:49 UTC from IEEE Xplore. Restrictions apply.
 generator G9 while disturbance (b) is considered to
excite interarea mode of oscillations of the system.
Without PSSs, the system responses due to
dlsturbances (a) and (b) are shown in Figs. 5 and 6
respectively. It is observed that the system damping is
very poor and the system is highly oscillatory.
Therefore, it is necessary to install stabilizers in order to
have good dynamic performance. To i d e n a the
optimum locations of PSSs, the participation factor
method [18] and the sensitivity of PSS effect (SPE)
method [19] were used. The results of both methods
indlcate that the generators G5, G7, and G9 are the
optimum locations for installing PSSs to damp out the
electromechanical modes of oscillations. Therefore,
these generators are equipped with three of the proposed
GFLPSSs.
The system responses with the proposed GFLPSSs
for disturbances (a) and (b) are shown in Figs. 7 and 8
respectively. In these Figs., the system performance
with the proposed GFLPSSs is compared with those
with CPSSs and FLPSSs [SI. The results illustrate the
Fig. 4: Single line diagram for New England system superiority of the proposed GFLPSSs to CPSSs and
FLPSSs and its efficiency to damp out the local modes
0.02 1 as well as the interarea modes of oscillations.
h

-
3
a
v
v)
c
0'01 1 h I\
.-0
, 0.00
I
.-m
a, ---
n
U ..............
a,
a, -
"a ----

0.00 1.oo 2.00 3.00 4.00 5.00

Time (s) - A09
Fig. 5: System response without PSSs for disturbance (a)
0.01 7

h
3
v
a
m
c
.-0
,
I
.-m 0.00
ar
e
U
a,

"2
-0.01

0.00
' I

1 .oo
l I

2.00
l

Time (s)
I

3.00
l [

4.00
l

5.00 4V
U
W
a,
"
-0.01

- A09
Fig. 6: System response without PSSs for disturbance (b) -002 I
I I
I I
I I
I
It is worth pointing out that disturbance (a) is
considered to excite local mode of oscillation of the

332

Authorized licensed use limited to: Sona College of Technology. Downloaded on August 02,2010 at 05:34:49 UTC from IEEE Xplore. Restrictions apply.
 0'01 1 h -
3
a
v
v)
.-
0 c
c
m --- Am5
0
G
.-(U
W ....... ....... > 0.00
n "O6 W
;-0.01 13

4I
Q,

, -0.02
0.00 2.00 4.00 6.00 8.00
U
a,
W
n
v)

-0.01 j 1
Time (s) 0.00 2.00 4.00 6.00 8.00
(c) Time (s)
Fig. 7: System response for disturbance (a) (b)
(a) with CPSSs
(b) with FLPSSs
(c) with proposed GFLPSSs

6. COORDINATION BETWEEN GFLPSS AND

CPSS

In most situations, the newly installed GFLPSSs

will have to work together with CPSSs which already
exist in a power system. In this section, system response
with the proposed GFLPSSs and CPSSs working
together has been also investigated. Several
combinations between the proposed GFLPSSs and
CPSSs are considered as shown in Figs. 9 and 10 where
-0.01 .a
0.00 2.00 4.00 6.00 8.00
C and P refer to CPSS and Proposed GFLPSS Time (s)
respectively. The first, second, and third letters in each (c)
combination denote the type of stabilizer installed on Fig. 8: System response for disturbance (b)
G5, G7, and G9 respectively. (a) with CPSSs
The system responses to the dsturbances (a) and (b) with FLPSSS
(b) with different combinations are shown in Figs. 9 and (c) with proposed GFLPSSs
10 respectively. It can be seen that the two types of PSSs
can work cooperatively. The response with the proposed 7. CONCLUSION
GFLPSSs and CPSSs combinations is better and the
oscillations are damped out much quicker than the In this study, a genetic-based fuzzy logic power
response with only CPSSs. Generally, the system system stabilizer is introduced. The proposed GFLPSS
performance is improved as the number of the proposed was designed by incorporating CA to search for the
GFLPSSs installed increases as shown in these Figures. optimal settings of FLPSS tuning parameters. The
--- simulation results show that the performance of FLPSS
A 9
0.01 7 .............. 6w6
can be improved significantly by incorporating m
genetic-based learning mechanism. It is shown that the
proposed GFLPSS can provide good damping
characteristics during transient conditions for local as
well as interarea modes of oscillations. In addition, the
coordination between the proposed GFLPSS and the
conventional stabilizers is demonstrated.

8. ACKNOWLEDGMENT
-0.01 I I I I 1
a.oo 2.00 4.00 6.00 8.00
Time (s) The authors would like to acknowledge the support
(a) of King Fahd University of Petroleum & Minerals.

333

Authorized licensed use limited to: Sona College of Technology. Downloaded on August 02,2010 at 05:34:49 UTC from IEEE Xplore. Restrictions apply.
 [7] M. Hassan, 0. P. Malik, and G. S. Hope, “A fuzzy
logic based stabilizer for a synchronous machine,”
IEEE Trans. EC, vol. 6, no. 3, 1991, pp. 407-413.
[SI T. Hiyama and T. Sameshma, “Fuzzy Logic
Control Scheme for On-Line Stabilization of
multimachme Power System,” Fuzzy Sets and
Systems, vol. 39, pp. 181-194, 1991.
----_ CCP Combtnabon [9] H. A. Toliyat, J. Sadeh, and R. Ghazi, “Design of
augmented fuzzy logic power system stabilizers to
--- PCP Combinatm enhance power system stability,” IEEE Trans. EC,
- PPP COmblMtIOn vol. 11, no. 1, 1996, pp. 97-103.
-002 [10]K. A. El-Metwally, G. C. Hancock, and 0. P.
0 00 2 00 4 00 6 00 8 00
Malik, “Implementation of a fuzzy logic PSS using
Time (5) a micro-controller and experimental test results, “

Fig. 9: System response with different combinations of IEEE Trans. EC, vol. 11, no. 1, 1996, pp. 91-96.
PSSs for dsturbance (a) [111 J. H. Holland, Adaptation in natural and artificial
systems, Addison-Wesley, 1975.
CCC Combinatin
[12]R. Dimeo and K. Y. Lee, “Boiler-turbine control
system design using a genetic algorithm,” IEEE
I _ _ _ -CPC
- Combination Trans. EC, vol. 10, no. 4, 1995, pp. 752-759.
[13]Y. L. Abdel-Magid and M. M. Dawoud, Genetic ”

algorithms applications in load frequency control,”

Conference of Genetic Algorithms in Engineering
Systems: Innovations and Applications, Sept. 1995,
pp, 207-213.
[14]P. Ju, E. Handschm, and F. Reyer, “Genetic
algorithm aided controller design with application
to SVC,” IEE Proc. Gen. Tran. Dist., vol. 143, no.
3, 1996, pp. 258-262.
-0.01 I [15]C.C. Karr and E.J. Gentry, “Fuzzy Control of pH
I
I I
I I I

0.00 2 00 4 00 600 8 00 Using Genetic Algorithms,” IEEE Trans. on Fuzzy

Time (s) Systems, vol. 1, no. 1, pp. 46-53, 1993.
Fig. 10: System response with different combinations of
[16]D. Park, A. Kandel, and G. Langholz, “Genetic-
PSSs for disturbance (b)
Based New Fuzzy Reasonoing Models with
Application to fuzzy Control,” IEEE Trans. syst.,
9. REFERENCES
Man, Cybern., vol. 24, no. 1, pp. 39-47, 1994.
[17]M. A. Abido and Y. L. Abdel-Magid, “Tuning of a
113 P. M. Anderson and A. A. Fouad, Power System fuzzy logic power system stabilizer using genetic
Control and Stability, Iowa State Univ. Press,
algorithms,”Accepted for Presentation in 4th IEEE
Ames, Iowa, U.S.A., 1977.
International Conference on Evolutionary
[2] Y. N. Yu, Electric Power System Dynamics,
Computation E E C ’97, Indianapolis, USA, April
Academic Press, 1983.
13-16, 1997.
[3] E. Larsen and D. Swann, “Applying power system
[lS]Y. Y. Hsu and C. L. Chen, “Identification of
stabilizers,” IEEE Trans. PAS, vol. 100, no. 6,
optimum location for stabilizer applications using
1981, pp. 3017-3046.
participation factors,” IEE Proc., pt. C, vol. 134,
[4] F. P. De Mello and T. F. Laskowski, “Concepts of
no. 3, May 1987, pp. 238-244.
power system dynamic stability,” IEEE Trans. PAS, [19]E. 2. Zhou, 0. P. Malik, and G. S. Hope, ”Theory
vol. 94, 1979, pp. 827-833.
and method for selection of power system stabilizer
[ 5 ] Y. Hsu and K. Liou, “Design of self-tuning PID location,” IEEE Trans. on Energy Conversion, vol.
power system stabilizers for synchronous
6, no. 1, 1991, pp. 170-176.
generators,” IEEE Trans. EC, vol. 2, no. 3, 1987,
pp. 343-348.
[6] D. Xia and G. T. Heydt, “Self-tuning controller for
generator excitation control,” IEEE Trans. PAS,
vol. 102, 1983, pp. 1877-1885.

334

Authorized licensed use limited to: Sona College of Technology. Downloaded on August 02,2010 at 05:34:49 UTC from IEEE Xplore. Restrictions apply.